2.0.3 - Azure Foundry support, SaaS trials and Telemetry with control

This release brings three things many of you have been asking for: Azure Foundry as an AI engine, transparent and controllable telemetry, and a self-service trial that gets you from sign-up to your first test run with no infrastructure to build. Azure Foundry support for OpenAI and Anthropic models ATP360 can now run its AI pipeline through your own Azure Foundry project, alongside the existing direct connections to OpenAI, Claude and Gemini. If your organisation already standardises on Azure for AI, you can now use the models deployed in your own project rather than connecting to each public provider separately. Connecting a project. Go to Settings > AI Integrations and open the Azure Foundry card. Paste the project endpoint URL from the Azure Foundry portal (it looks like https://.services.ai.azure.com/api/projects/) along with your API key. As with the other engines, ATP360 tests the connection before saving anything. Choosing deployments. Once connected, every deployment in the project is listed and you tick the ones you want ATP360 to use. The deployment name is what Azure receives as the model on each request. This release supports the OpenAI and Anthropic models we have tested. Anything else in your project, such as DeepSeek, Kimi or Grok deployments, is still shown but marked Not Supported for now. Computer use is checked for you. Not every model can drive a desktop. When you add a deployment, ATP360 checks whether its model supports computer use. Only deployments that pass that check can be used at the Execution layer. Any supported deployment can be used for parsing and orchestration. Mix and match across the pipeline. Foundry deployments can be combined freely with the other engines. In the example below, a Claude Opus deployment in Azure Foundry handles text parsing and orchestration, Gemini parses video, and OpenAI drives execution. Changes on the Test Now screen apply to that run only. Workspace defaults stay in Settings. A note on cost reporting. ATP360 calculates the cost of a run from the tokens used and the pricing it holds for each model. If a model has no known price, you'll see a warning and the run will report a $0 cost. The test still runs normally. Telemetry: you decide what's shared ATP360 can now report facts about completed test runs back to WorkspaceDNA. We use them to compare AI engines on efficacy, cost and override rate, and to act on what we learn. That means retiring models that underperform, tuning the ones that do well, and bringing down the cost of a run for everyone. You control how much is sent from the new Telemetry tab in Settings. Changes apply immediately. There are three levels: Off - Nothing is reported. No run facts leave your tenant. Basic (analytics)Structured run facts only: identifiers, outcomes, timings, token counts and costs. No free text. Advanced (diagnostics)Everything in Basic, plus the note a reviewer types when overriding an AI verdict. Also grants permission for error messages and diagnostic bundles in future releases. Advanced is the default for new workspaces. Full details are in the ATP360 Administration Guide. Self-service trial with SaaS deployment You can now sign up for an ATP360 trial and start testing without deploying anything yourself. At sign-up you choose how you want to run ATP360: SaaS: we host ATP360 for you. Your workspace is provisioned automatically, so you can go straight to connecting a CUA worker and running your first test. PaaS: deploy ATP360 into your own Azure environment, as before. SaaS trials come with $25 of AI credits, so you can run real tests without connecting your own AI engine keys first. Each trial lasts 30 days. Get started with your trial here